US 12,292,920 B2
Geo-visual search
Ryan S. Keisler, Oakland, CA (US); Samuel W. Skillman, Santa Fe, NM (US); and Michael S. Warren, Los Alamos, NM (US)
Assigned to EarthDaily Analytics USA, Inc., Maple Grove, MN (US)
Filed by Descartes Labs, Inc., Santa Fe, NM (US)
Filed on May 6, 2022, as Appl. No. 17/738,433.
Application 17/738,433 is a continuation of application No. 16/285,077, filed on Feb. 25, 2019, granted, now 11,354,352.
Application 16/285,077 is a continuation of application No. 15/497,598, filed on Apr. 26, 2017, granted, now 10,248,663, issued on Apr. 2, 2019.
Claims priority of provisional application 62/466,588, filed on Mar. 3, 2017.
Prior Publication US 2022/0391437 A1, Dec. 8, 2022
Int. Cl. G06F 16/583 (2019.01); G06F 16/29 (2019.01); G06F 16/687 (2019.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01); G06V 20/00 (2022.01)
CPC G06F 16/583 (2019.01) [G06F 16/29 (2019.01); G06F 16/5854 (2019.01); G06F 16/687 (2019.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06V 20/35 (2022.01)] 17 Claims
OG exemplary drawing
 
1. A system, comprising:
one or more processors configured to:
receive a set of exemplar locations;
for an exemplar location in the set of exemplar locations:
determine an identifier of a tile corresponding to the exemplar location; and
determine an exemplar feature vector corresponding to the identifier of the tile, wherein determining the exemplar feature vector comprises extracting visual features from the tile at least in part by performing feature extraction on spectral data associated with pixels of the tile;
for the exemplar feature vector, determine a set of visual neighbor feature vectors corresponding to tiles determined to be visual neighbors of the tile corresponding to the exemplar location, wherein inclusion of a visual neighbor feature vector in the set of visual neighbor feature vectors is based at least in part on determining a hamming distance between the visual neighbor feature vector and the exemplar feature vector; and
associate the exemplar feature vector with the set of visual neighbor feature vectors at least in part by storing a set of files comprising:
identifiers of the tiles determined to be visual neighbors of the tile corresponding to the exemplar location; and
the set of visual neighbor feature vectors corresponding to the identifiers of the tiles determined to be visual neighbors, including the visual neighbor feature vector that is included based at least in part on the determining of the hamming distance; and
wherein at query time, a search involving a query image tile is facilitated at least in part by:
performing a first comparison between a query feature vector corresponding to the query image tile and the exemplar feature vector;
based at least in part on the first comparison between the query feature vector and the exemplar feature vector, performing a second comparison between the query feature vector and visual neighbor feature vectors in the stored set of visual neighbor feature vectors determined for the exemplar feature vector; and
based at least in part on the second comparison, returning, as output, a tile identifier of the visual neighbor feature vector included in the stored set of visual neighbor feature vectors determined for the exemplar feature vector; and
a memory coupled to the one or more processors and configured to provide the one or more processors with instructions.